Predictive analytics is the practice of learning from historical data in contemplation of making decisions about the future. Predictive analytics is not new to healthcare. However, in the past it has been limited by many factors, including data availability and accessibility. Over the last few years, the issue of data availability has been rectified, mainly due to the widespread implementation of electronic health records (EHRs) and the growth of patient generated data. But in order to leverage this data, we still need to analyze it and reach actionable insights. Here is where healthcare data analytics companies come in.
Even though many giants, like IBM, Oracle, and McKesson, are heavily involved in this market, startups play a vital role in moving this industry forward. One such startup is Medial EarlySign, an Israeli company that utilizes machine learning and artificial intelligence to build predictive analytics algorithms. These algorithms are structured to help improve healthcare quality while simultaneously controlling costs. We at Medgadget had the chance to interview Mr. Ori Geva, Co-Founder and CEO of Medial EarlySign, to learn more about the company.
Kenan Raddawi, MD, Medgadget: Could you please give our readers an overview of Medial EarlySign, its mission, and main products?
Ori Geva: Medial EarlySign’s machine learning-based solutions empower healthcare organizations to identify patients with a high probability for harboring or developing specific illnesses – including cancers, diabetes and other life-threatening conditions – by leveraging blood test results and routine electronic health record (EHR) data. We augment the value of existing clinical and lab data and refine risk management processes to create actionable opportunities to intervene early and delay or prevent progression of illness more effectively, helping to improve patient outcomes and focus financial resources. Our technology is supported by real results in clinical practice, as well as peer-reviewed research published by internationally recognized health organizations and hospitals, including Oxford University and Kaiser Permanente.
Medgadget: Can you tell us more about some of the aspects that Medial EarlySign’s solutions assist with, and give us a real life example of a solution that has proven to be effective?
Ori Geva: We at MES are addressing the challenges our customers face. We identify different decision-making processes or health management activities that could be improved with additional data or insights. Based on that, we configure or customize our models or create new ones to help our customers make improved decisions, using the insights from these models to improve their outcomes. This could be by having more fine-grained stratification, providing “safety-net” hints, enhancing the predictive value of known lab tests, or mediating information on behalf of our customers to their members with the goal of encouraging greater participation. In each specific domain, we strive to ameliorate the knowledge at hand and leverage the ability of machine learning to process huge amounts of data, finding subtle patterns and relations that are difficult for humans to discern or too complicated to be captured in a simple formula.
One example is a solution that has been implemented in clinical practice for more than two years in both the Israeli and EU healthcare systems. This solution helps health systems to invest additional resources to reach out to individuals who do not comply with recommendations and screening guidelines and may be at high risk for lower GI disorders (including colorectal cancer). Our solutions work in the background, and when a non-compliant individual performs a routine blood count test, the model analyzes these results, combining it with their age and sex, to flag if they are at high risk for having GI disorders and whether the health system should invest additional resources to have this person examined. To date, our system has flagged over 100 individuals who were later diagnosed by colonoscopy as having colorectal cancer or precancerous conditions. Many of these are individuals whose cancers may not have been discovered until much later.
Medgadget: As we all know, data in healthcare can be messy. What are the types and sources of data that Medial EarlySign relies on, and how do you go about cleaning up this data before using it in building your models?
Ori Geva: We rely on data sources emanating from various healthcare organizations in several geographies around the world. Our expertise lies in routine medical data, such as lab test results. This data commonly sits dormant at many healthcare organizations. The idea behind our data collection methodology is to collect a diverse set of data so that we can compare and validate our models on diverse populations. We have a rich set of tools to analyze the data and assess whether it represents the populations in these regions. We also have the comparison tools necessary to find any anomalies in the data and compare them to previous data sets.
Medgadget: How do you make sure that the models you build would be effective on a wide variety of patient populations and healthcare environments?
Ori Geva: As mentioned above, we collect data from several geographies and so can validate the models on various populations in-house. However, that is only the beginning of the process. Many of our models also go through external validations thanks to our collaborations with leading healthcare institutions. For example, clinical data studies validating our lower GI disorder models were conducted in collaboration with Oxford University for U.K. populations, and with Kaiser Permanente for U.S. populations. These studies were published and can be downloaded here. Our environment also allows local validations with health systems if such are requested.
Medgadget: In the last few years, there has been an explosion in the number of healthcare analytics startups. Could you tell us what sets Medial EarlySign apart?
Ori Geva: Firstly, validation. We have found (understandably) that most healthcare organizations won’t implement any models until they see evidence that it works.
Secondly, studies proving the effectiveness of our models in clinical practice. Very few companies have actual measurable prospective results from clinical practice implementations.
Thirdly, we have global partnerships with 11 leading institutions around the world, including the U.K. and China, and three of the top five U.S. Healthcare providers. This is vital to building dependable predictive models that organizations can rely on and actually want to implement.
Fourthly, we are fortunate to have an exceptional scientific advisory board consisting of established leaders in medicine, genetics, population health and digital health.
Finally, we are focusing on clinical and intervention oriented modelling, optimizing for the decision our customers make and the outcomes they measure, it is not merely about risk adjustment, cost adjustment or enlarging registries.
Medgadget: The market of healthcare data analytics has been growing rapidly over the last few years. How do you foresee the future of this market? Do you think that just a few big companies will dominate it? And how would small and medium-sized companies maintain their competitive edge?
Ori Geva: The industry is young and it’s hard to make predictions. However, when the dust settles, it is more likely than not that there will be several large companies who end up becoming the market leaders. That said, some of these companies will likely build their solutions through collaborations and acquisitions of smaller, more specialized companies. So, we expect to see consolidation. The way to thrive (rather than just to survive), is to continue to focus on what we do and be the best at it. It will take any company, whether large or small, time to reach our clinical and technological milestones since in healthcare proof takes time.
